Android的Audio音頻系統(tǒng)

1、音頻系統(tǒng)JJJNNNIII初始化在 system_init（運行在 Simulator 上）或者 main_Mediaserver 中，AudioFlinger 被創(chuàng)建，會生成一個 AudioHardwareInterface 實例（Android 定義的音頻設備的一個抽象層），并且初始化音頻系統(tǒng)的模式和路由信息如下：mHardwareStatus = AUDIO_HW_IDLE; mAudioHardware = AudioHardwareInterface:create(); mHardwareStatus = AUDIO_HW_INIT;if (mAudioHardware-initCh

2、eck() = NO_ERROR) / open 16-bit output stream for s/w mixer mHardwareStatus = AUDIO_HW_OUTPUT_OPEN;mOutput = mAudioHardware-openOutputStream(AudioSystem:PCM_16_BIT); mHardwareStatus = AUDIO_HW_IDLE;if (mOutput) mSampleRate = mOutput-sampleRate();mChannelCount = mOutput-channelCount();mFormat = mOutp

3、ut-format();mMixBufferSize = mOutput-bufferSize();mFrameCount = mMixBufferSize / mChannelCount / sizeof(int16_t); mMixBuffer = new int16_tmFrameCount * mChannelCount; memset(mMixBuffer, 0, mMixBufferSize);mAudioMixer = new AudioMixer(mFrameCount, mSampleRate);/ FIXME - this should come from settings

4、 setMasterVolume(1.0f);setRouting(AudioSystem:MODE_NORMAL, AudioSystem:ROUTE_SPEAKER,AudioSystem:ROUTE_ALL);setRouting(AudioSystem:MODE_RINGTONE,AudioSystem:ROUTE_SPEAKER, AudioSystem:ROUTE_ALL); setRouting(AudioSystem:MODE_IN_CALL, AudioSystem:ROUTE_EARPIECE,AudioSystem:ROUTE_ALL); setMode(AudioSys

5、tem:MODE_NORMAL); mMasterMute = false; else LOGE(Failed to initialize output stream); else LOGE(Couldnt even initialize the stubbed audio hardware!);在 SystemServer 啟動的時候，會生成一個 AudioService 的實例， try Log.i(TAG, Starting Audio Service);ServiceManager.addService(Context.AUDIO_SERVICE, new AudioService(c

6、ontext); catch (Throwable e) Log.e(TAG, Failure starting Volume Service, e);AudioService 的構造函數會讀取一些關于音頻的配置信息，比如 Ringer 和 vibrate 信息，private void readPersistedSettings() final ContentResolver cr = mContentResolver;mRingerMode=System.getInt(cr,System.MODE_RINGER,AudioManager.RINGER_MODE_NORMAL);mRinge

7、rModeAffectedStreams = System.getInt(mContentResolver,System.MODE_RINGER_STREAMS_AFFECTED,1AudioSystem.STREAM_RING);mVibrateSetting = System.getInt(cr, System.VIBRATE_ON, 0);mMuteAffectedStreams = System.getInt(cr,System.MUTE_STREAMS_AFFECTED,(1 AudioSystem.STREAM_MUSIC)|(1 AudioSystem.STREAM_RING)|

8、(1 AudioSystem.STREAM_SYSTEM);/ Each stream will read its own persisted settings/ Broadcast the sticky intent broadcastRingerMode();/ Broadcast vibrate settings broadcastVibrateSetting(AudioManager.VIBRATE_TYPE_RINGER); broadcastVibrateSetting(AudioManager.VIBRATE_TYPE_NOTIFICATION);同時也會從底層音頻系統(tǒng)讀取模式和

9、路由信息：private void readAudioSettings() synchronized (mSettingsLock) mMicMute = AudioSystem.isMicrophoneMuted();mMode = AudioSystem.getMode();for (int mode = 0; mode AudioSystem.NUM_MODES; mode+) mRoutesmode = AudioSystem.getRouting(mode);在 AudioSystem.cpp 第一次調用 get_audio_flinger 成功后，它會通過 binder 來監(jiān)聽運行

10、在 media_server 進程中的 AudioFlinger 是否活著。/ establish binder interface to AudioFlinger serviceconst sp& AudioSystem:get_audio_flinger()Mutex:Autolock _l(gLock); if (gAudioFlinger.get() = 0) sp sm = defaultServiceManager(); sp binder;do binder = sm-getService(String16(media.audio_flinger); if (binder !=

11、0)break;LOGW(AudioFlinger not published, waiting.);usleep(500000); / 0.5 s while(true);if (gDeathNotifier = NULL) gDeathNotifier = new DeathNotifier(); else if (gAudioErrorCallback) gAudioErrorCallback(NO_ERROR);binder-linkToDeath(gDeathNotifier);gAudioFlinger = interface_cast(binder);LOGE_IF(gAudio

12、Flinger=0, no AudioFlinger!?);return gAudioFlinger;到此，整個音頻系統(tǒng)初始化完畢。重新啟動如果 AudioFlinger 運行的 media_server 進程異常死掉，AudioSystem 會收到一個事件通知， void AudioSystem:DeathNotifier:binderDied(const wp& who) Mutex:Autolock _l(AudioSystem:gLock);AudioSystem:gAudioFlinger.clear();if (gAudioErrorCallback) gAudioErrorCal

13、lback(DEAD_OBJECT);LOGW(AudioFlinger server died!);從而調用 android_media_AudioSystem.cpp 注冊下來的回調函數，該函數又是通過 JNI 來調用 AudioService.java 注冊下來的回調函數，在該函數中會發(fā)送MSG_MEDIA_SERVER_DIED消息，AudioService 會監(jiān)聽這個消息，這樣 AudioService 就能知道 AudioFlinger 已不工作，它就接著調用 getMode 來嘗試連接到重啟后的 AudioFlinger。case MSG_MEDIA_SERVER_DIED:Lo

14、g.e(TAG, Media server died.);/ Force creation of new IAudioflinger interface mMediaServerOk = false; AudioSystem.getMode();break;當連接成功后，AudioFlinger 會調用 android_media_AudioSystem.cpp 注冊下來的回調函數，該函數又是通過 JNI 來調用 AudioService.java 注冊下來的回調函數，在該函數中會發(fā)送 MSG_MEDIA_SERVER_STARTED 消息。接著 AudioService 就去配置底層音頻系統(tǒng)

15、，包括模式、路由、每一路流的音量大小和 Ringer 狀態(tài)。case MSG_MEDIA_SERVER_STARTED:/ Restore audio routing and stream volumes applyAudioSettings();for (int streamType = AudioSystem.NUM_STREAMS - 1; streamType= 0; streamType-) int volume;VolumeStreamState streamState = mStreamStatesstreamType; if (streamState.muteCount() =

16、 0) volume = streamState.mVolumesstreamState.mIndex; else volume = streamState.mVolumes0;AudioSystem.setVolume(streamType, volume);setRingerMode();Note: AudioSystem 的 Native 實 現 在 device/libs/android_runtime/android_media_AudioSystem.cpp 中。模式初始的時候音頻系統(tǒng)是處于 MODE_NORMAL 模式的，下面是其模式狀態(tài)變遷圖：問題：當一個 Ringtone 放

17、完了之后，理論上系統(tǒng)是否要自動切換回 NORMAL 模式而不是必須要主動調用 stopRing？我沒找到相關 code。路由信息1. 當 HeadsetObserver 檢測到有耳機插上來的時候，它會把音頻系統(tǒng)的路由設置成均使用該耳機；當耳機被拔下來后，它會把音頻系統(tǒng)的路由設置成缺省配置（即都通過揚聲器）。private synchronized final void update(String newName, int newState) if (newName != mHeadsetName | newState != mHeadsetState) mHeadsetName = newNa

18、me; mHeadsetState = newState;AudioManager audioManager = (AudioManager) mContext.getSystemService(Context.AUDIO_SERVICE);if (mHeadsetState = 1) audioManager.setRouting(AudioManager.MODE_NORMAL,AudioManager.ROUTE_HEADSET,AudioManager.ROUTE_ALL);audioManager.setRouting(AudioManager.MODE_RINGTONE,Audio

19、Manager.ROUTE_HEADSET|AudioManager.ROUTE_SPEAKER,AudioManager.ROUTE_ALL);audioManager.setRouting(AudioManager.MODE_IN_CALL, AudioManager.ROUTE_HEADSET,AudioManager.ROUTE_ALL); else audioManager.setRouting(AudioManager.MODE_NORMAL,AudioManager.ROUTE_SPEAKER,AudioManager.ROUTE_ALL);audioManager.setRou

20、ting(AudioManager.MODE_RINGTONE, AudioManager.ROUTE_SPEAKER,AudioManager.ROUTE_ALL);audioManager.setRouting(AudioManager.MODE_IN_CALL, AudioManager.ROUTE_EARPIECE,AudioManager.ROUTE_ALL);sendIntent();PhoneApp 會接收到一個 ACTION_HEADSET_PLUG 的 Intent，往自身發(fā)送一個消息，處理如下：只是在沒有藍牙耳機或者未使用藍牙耳機，而且該有線耳機是被拔掉的情況下，才把路由信

21、息設置成 MODE_IN_CALL 走 ROUTE_SPEAKER。 case EVENT_WIRED_HEADSET_PLUG:/ Since the presence of a wired headset or bluetooth affects the/ speakerphone, update the speaker state. We ONLY want to do/ this on the wired headset connect / disconnect events for now/ though,sowereonlytriggeringonEVENT_WIRED_HEADS

22、ET_PLUG.if (!isHeadsetPlugged() &(mBtHandsfree = null | !mBtHandsfree.isAudioOn() / is the state is not connected, restore the speaker state. PhoneUtils.restoreSpeakerMode(getApplicationContext();NotificationMgr.getDefault().updateSpeakerNotification(); break;問題：假設耳機插上來之前是通過藍牙耳機在接聽電話（或者聽音樂）的，耳機插上來的時

23、候系統(tǒng)就自動切換到使用耳機了，但是 PhoneApp 這個時候并不知道，它還以為在繼續(xù)使用藍牙耳機。但當耳機拔掉之后，怎么再切換回繼續(xù)使用藍牙耳機呢（系統(tǒng)默認是切換成EARPIECE）？這個時候 PhoneApp 的狀態(tài)應該是不對的。2. 當在 Setting 里面把藍牙耳機配對和 RFCOMM 連接上之后，BluetoothHeadsetService 會負責去和藍牙耳機建立 SCO 連接，當連接完成之后 BluetoothHandsfree 會調用AudioManager 的 setBluetoothScoOn 函數來通知音頻系統(tǒng)去切換 MODE_IN_CALL 路由信息到使用 ROUTE

24、_BLUETOOTH。/* Sets audio routing to the Bluetooth headset on or off.* param on set true to route SCO (voice) audio to/from Bluetooth* headset; false to route audio to/from phone earpiece*/public void setBluetoothScoOn(boolean on)setRouting(MODE_IN_CALL, on ? ROUTE_BLUETOOTH : ROUTE_EARPIECE, ROUTE_A

25、LL);當藍牙設備被關閉或者鏈接斷掉的時候，BluetoothHeadsetService 會收到一個 DISABLED_ACTION 的 Intent ，接著 BluetoothHandsfree 會調用 AudioManager 的 setBluetoothScoOn 函數來通知音頻系統(tǒng)去切換 MODE_IN_CALL 路由信息到ROUTE_EARPIECE。問題：setBluetoothScoOn 的實現在處理藍牙設備被關閉的時候，是直接把路由信息改成到ROUTE_EARPIECE，并沒有恢復到使用藍牙設備之前的信息狀態(tài)。Ringtone 是否需要在藍牙耳機上播放呢？音量控制對外接口是

26、AudioManager播放音頻系統(tǒng)對外的播放接口是 AudioTrack，每一路音頻會對應一個 AudioTrack 的實例，它會通過 iBinder 來遠程調用 AudioFlinger 的 createTrack 函數。/ create the tracksp track = audioFlinger-createTrack(getpid(),streamType, sampleRate, format, channelCount, bufferCount, flags);if (track = 0) LOGE(AudioFlinger could not create track);r

27、eturn NO_INIT;sp cblk = track-getCblk();if (cblk = 0) LOGE(Could not get control block);return NO_INIT;而 AudioFlinger 的 createTrack 又會在內部生成一個 Track 實例，再將其包裝成 TrackHandle 返回給 AudioTrack。track = new Track(this, client, streamType, sampleRate, format,channelCount,bufferCount,channelCount=1?mMixBufferSi

28、ze1:mMixBufferSize);mTracks.add(track);trackHandle = new TrackHandle(track);return trackHandle;所以 AudioTrack 和 AudioFlinger 實際操作的都是 Track 實例，AudioTrack 通過它來執(zhí)行控制操作（start/stop）和寫入操作（write），AudioFlinger 則負責管理多個 Track（包括調用 AudioMixer 來混音）。兩者之間的關系可以用生產者/消費者來類比，AudioTrack 是生產者， AudioFlinger 則是消費者。AudioTra

29、ckAudioTrack 的 start/stop 操作可以理解成一個開關，控制的是是否將與之對應的 Track 實例納入 AudioFlinger 的管理中去，下面僅以 start 操作為例。 void AudioTrack:start()LOGV(start);if (mAudioTrackThread != 0) mAudioTrackThread-mLock.lock();if (android_atomic_or(1, &mActive) = 0) setpriority(PRIO_PROCESS, 0, THREAD_PRIORITY_AUDIO_CLIENT);mActive =

30、 1;mAudioTrack-start();if (mAudioTrackThread != 0) mAudioTrackThread-run(AudioTrackThread,THREAD_PRIORITY_AUDIO_CLIENT);if (mAudioTrackThread != 0) mAudioTrackThread-mLock.unlock();status_t AudioFlinger:TrackHandle:start() return mTrack-start();status_t AudioFlinger:Track:start()LOGV(start(%d), mNam

31、e);mAudioFlinger-addTrack(this);return NO_ERROR;status_t AudioFlinger:addTrack(const sp& track)Mutex:Autolock _l(mLock);/ here the track could be either new, or restarted/ in both cases unstop the trackif (track-isPaused() track-mState = TrackBase:RESUMING; LOGV(PAUSED = RESUMING (%d), track-name();

32、 else track-mState = TrackBase:ACTIVE;LOGV(? = ACTIVE (%d), track-name();LOGV(mWaitWorkCV.broadcast);mWaitWorkCV.broadcast();if (mActiveTracks.indexOf(track) mFillingUpStatus = Track:FS_FILLING;mActiveTracks.add(track);return NO_ERROR;return ALREADY_EXISTS;AudioTrack 的 write 則是往 audio_track_cblk_t 結

33、構中寫入數據。ssize_t AudioTrack:write(const void* buffer, size_t userSize)LOGV(write %d bytes, mActive=%d, userSize, mActive);ssize_t written = 0;do if (mPosition = 0) status_t err = obtainBuffer(&mAudioBuffer, true);if (err 0) / out of buffers, return #bytes writtenif (err = status_t(NO_MORE_BUFFERS)brea

34、k;return ssize_t(err);size_t capacity = mAudioBuffer.size - mPosition;size_t toWrite = userSize capacity ? userSize : capacity;memcpy(mAudioBuffer.i8 + mPosition, buffer, toWrite);buffer = static_cast(buffer) + toWrite;mPosition += toWrite;userSize -= toWrite;capacity -= toWrite;written += toWrite;i

35、f (capacity = 0) mPosition = 0;releaseBuffer(&mAudioBuffer); while (userSize); return written;AudioFlingerAudioFlinger 對 Track 的管理是實現在 threadLoop 中的。先檢測進入 standby 的超時是否到了，超時的話 AudioFlinger 會調用 AudioHardwareInterface 的 standby，這個是為省電考慮的。nsecs_t standbyTime = systemTime();do enabledTracks = 0; / scope

36、 for the lock Mutex:Autolock _l(mLock);const SortedVector wp & activeTracks = mActiveTracks; / put audio hardware into standby after short delayif UNLIKELY(systemTime() standbyTime) / wait until we have something to do.LOGD(Audio hardware entering standbyn); mHardwareStatus = AUDIO_HW_STANDBY; if (!

37、mStandby) mAudioHardware-standby(); mStandby = true;mHardwareStatus = AUDIO_HW_IDLE;/ were about to wait, flush the binder command buffer IPCThreadState:self()-flushCommands(); mWaitWorkCV.wait(mLock);LOGD(Audio hardware exiting standbyn); standbyTime = systemTime() + kStandbyTimeInNsecs;continue;如果

38、未進 standby，接下來遍歷所有當前 Active 的 Track 實例。/ find out which tracks need to be processed size_t count = activeTracks.size();for (size_t i=0 ; icount ; i+) sp t = activeTmote(); if (t = 0) continue;Track* const track = t.get();audio_track_cblk_t* cblk = track-cblk();uint32_t u = cblk-user;uint32

39、_t s = cblk-server;/ The first time a track is added we wait/ for all its buffers to be filled before processing it audioMixer().setActiveTrack(track-name();當有某個 Track 的數據需要處理時（數據存儲在 audio_track_cblk_t 結構中，其 user 域表明當前寫入指針在 buffer 中的位置，server 域表明讀取指針在 buffer 中的位置，所以只有當 user 大于 server 的時候說明有數據要處理），先計

40、算該 Track 的 volume 信息，然后去配置針對該路 Track 的 Mixer 信息。if (u s) & (track-isReady(u, s) | track-isStopped() & !track-isPaused()/ compute volume for this track/ setup mixer needed information here AudioMixer& mixer(audioMixer(); mixer.setBufferProvider(track); mixer.enable(AudioMixer:MIXING);enabledTracks+;

41、else 最后進入真正的混音操作，再把混音過后的數據寫到 AudioHardwareInterface 生成的AudioOutputStream 中，由此整個音頻輸出完成。if (LIKELY(enabledTracks) / mix buffers.audioMixer().process(curBuf);/ output audio to hardware mLastWriteTime = systemTime(); mInWrite = true; mOutput-write(curBuf, mixBufferSize); mNumWrites+;mInWrite = false;mSt

42、andby = false;nsecs_t temp = systemTime();standbyTime = temp + kStandbyTimeInNsecs;nsecs_t delta = temp - mLastWriteTime;if (delta maxPeriod) LOGW(write blocked for %llu msecs, ns2ms(delta); mNumDelayedWrites+;sleepTime = kBufferRecoveryInUsecs; else 當所有 Active 的 Track 都沒有數據需要處理的時候，AudioFlinger 會 us

43、leep 一段時間從而進入 standby。/ There was nothing to mix this round, which means all/ active tracks were late. Sleep a little bit to give/ them another chance. If were too late, the audio/ hardware will zero-fill for us.LOGV(no buffers - usleep(%lu), sleepTime);usleep(sleepTime);if (sleepTime kMaxBufferReco

44、veryInUsecs) sleepTime += kBufferRecoveryInUsecs;錄制音頻系統(tǒng)對外的錄制接口是 AudioRecord，它會通過 iBinder 來遠程調用 AudioFlinger 的 openRecord 函數。/ open record channelsp record = audioFlinger-openRecord(getpid(), streamType, sampleRate, format, channelCount, bufferCount, flags);if (record = 0) return NO_INIT;sp cblk = re

45、cord-getCblk();if (cblk = 0) return NO_INIT;if (cbf != 0) mClientRecordThread = new ClientRecordThread(*this); if (mClientRecordThread = 0) return NO_INIT;而 AudioFlinger 的 openRecord 又會在內部先生成一個 AudioRecordThread 并且拿到AudioStreamIn，/ Create audio thread - take mutex to prevent race conditionMutex:Auto

46、lock _l(mLock);if (mAudioRecordThread != 0) LOGE(Record channel already open); goto Exit;thread = new AudioRecordThread(this); mAudioRecordThread = thread;/ Its safe to release the mutex here since the client doesnt get a/ handle until we return from this call/ open driver, initialize h/winput = mAu

47、dioHardware-openInputStream(AudioSystem:PCM_16_BIT, channelCount, sampleRate);再生成一個 RecordTrack 實例，將其包裝成 RecordTrackHandle 返回給 AudioRecord。 / create new record track and pass to record threadrecordTrack = new RecordTrack(this, client, streamType, sampleRate, format, channelCount, bufferCount, input-bufferSize();/ spin up record threadthread-open(recordTrack, input);thread-run(AudioRecordThread, PRIORITY_URGENT_AUDIO);/ return to handle to clientrecordHandle = new RecordHandle(recordTrack);所以 AudioRecord 和 AudioFlinger